Tair AI Assistant

Intelligent Q&A

Features

Provides a natural language-driven intelligent Q&A service based on official Alibaba Cloud documentation and the Tair expert knowledge base. You can use @ to select a specific instance for more targeted answers.

Use cases

• Product feature inquiries: Ask about product-related information for a Tair instance, such as its architecture type, version differences, and parameter configurations.

• Best practice guidance: Get expert recommendations for O&M scenarios, such as large key governance, hot key optimization, memory management, and connection pool configuration.

• Troubleshooting guidance: Get solutions for common issues such as connection timeouts, high memory usage, and latency jitter.

• Operational guidance: Get instructions for operations like creating an instance, configuring a whitelist, setting up data synchronization, and enabling an audit log.

Intelligent inspection

Features

This feature lets you perform one-click intelligent inspections. You can select inspection items and a time range, and the system automatically collects instance runtime data to generate an inspection report, helping you proactively identify potential risks.

Inspection items

Inspection item details

Data node performance

This inspection pinpoints performance bottlenecks by analyzing latency insight data from DB nodes across two dimensions: commands and events.

• Command dimension: Analyzes the maximum response time (maxRT) and number of calls for high-latency commands on each DB node to identify whether these commands are concentrated on a specific node. Examples include high-complexity commands such as HGETALL, KEYS, and SMEMBERS.

• Event dimension: Detects system-level events that affect performance, such as fork (creates a child process for RDB persistence or AOF rewrite), active-defrag-cycle (memory defragmentation), expire-cycle (expired key cleanup), aof-write (AOF write), and eviction-cycle (data eviction), and provides corresponding optimization suggestions.

Proxy node performance

By comparing the response time (RT) data between each Proxy and DB node, the system identifies the following four abnormal patterns:

It also checks whether multiple abnormal nodes are located on the same physical host to pinpoint host-level performance issues.

Slow log analysis

This inspection analyzes the slow query logs from both DB nodes and Proxy nodes:

• DB slow log: Aggregates and analyzes slow command details from each node, including the command type, execution count (cnt), maximum response time (maxRT), and source IP distribution. This feature focuses on identifying high-risk command patterns, such as KEYS full scans (use SCAN instead), large-range ZRANGEBYSCORE queries (narrow the query range or optimize indexes), and high-frequency EVAL (Lua script).

• Proxy slow log: Analyzes the distribution of slow requests across DB nodes, client sources, and command types from the perspective of the Proxy, allowing you to identify DB nodes where slow requests are concentrated and determine the root cause by considering the time complexity of commands (for example, if both O(1) and O(N) commands slow down simultaneously, the slowdown is likely caused by blocking from the O(N) commands).

Large key/hot key

This analysis covers the following four dimensions:

The inspection report analyzes the distribution of large/hot keys across nodes, paying special attention to data skew caused by hashtags (such as {tag}), and recommends investigating related calls by cross-referencing them with slow logs.

Events and alerts

This inspection aggregates O&M events and alerts within the specified time range, including failovers, kernel upgrades, resource warnings, and high-traffic requests. The high-traffic analysis identifies concentrated command patterns, client IP sources, and target DB node distributions to find the root cause of throttling triggers.

Time range

The inspection supports the following time range options:

• Last 1 hour

• Last 3 hours

• Last 24 hours

• Custom time range: Specify a custom start and end time. The maximum duration is 24 hours by default, and you can inspect data from up to 7 days prior.

Procedure

1. In the Tair AI Assistant panel, select the target instance to inspect.

2. Select the required inspection items. You can also select all items.

3. Select the time range for the inspection.

4. Click Start Inspection and wait for the inspection report to be generated.

5. View the inspection results and apply optimizations based on the recommendations.